HMM-Based Automatic Arrangement for Guitars with Transposition and its Implementation

Automatic generation of guitar tablatures for given songs has been one of research interests in music information processing. Furthermore, some of recent studies have attempted automatic arrangement of given songs before tablature generation for producing guitar scores for songs composed for other instruments. In our previous work, we have formulated “fingering decision” and “arrangement” in a unified framework based on hidden Markov model (HMM) whose hidden states are the left hand forms on the fingerboard and an observed sequence is a note sequence of a given song. The purpose of the present paper is to extend the HMM-based automatic arrangement to “arrangement with transposition” and introduce a web application that implements the arrangement with transposition. The optimal transposition for arrangement of a given song is obtained through a full search for all the possible keys where the resulting arrangements are evaluated based on the probabilities of the sequences of the left hand forms.